In-editor terminal for VSCode
Use a regular text tab in vscode as your terminal. Use the same extensions and shortcuts between your text editing and terminal usage.
Inspired by Bashbook which does roughly the same thing but instead uses a notebook view within vscode.
Why?
I use voice control to code half the time to relieve stress from my hands. It is possible due to talonvoice.com + cursorless.org which allows me to effectively move around in vscode editor. Unfortunately it is not available in terminal.
How to use
- Run it from command pallet
Bash Editor: Open new
- This will open a new tab, simply type any bash command for example
ls
- Press enter to run the command as in normal terminal
- Command's output will be appended to the same editor
Features and shortcomings
- No third party dependencies - vscode + node apis only
- Keep the tail of the terminal always visible
- Does not support interactivity yet (for example if the comment your running requires input)
- Does not support colors yet (will attempt to strip down colors and other special symbols from command output)
- Does not have autocomplete for paths, or anything else for that matter
- Does not have history yet
See Todo four things I'm considering to add.
Publishing
Publisher url https://marketplace.visualstudio.com/manage/publishers/bra1ndump
Run vsce publish from the project root.
It will ask for authentication token which you can get from azure here https://dev.azure.com/bra1ndump/_usersSettings/tokens. You can't access as the old one, just generate the new one (look at the scopes requested on the old one. Market Place > Manage).
To Do
🥼 - Effort units
📈 - Impact
- Bug, TextEditor#edit not possible on closed editors. Stops being able to write to editor on unfocus it seems
- TextEditor is closed/disposed??
- Need to create temporary file every time. Instead save with a new extension like in-editor-terminal.md. If already in directory simply open the existing one. If opened by other means - start listingin to edits
- [🥼🥼 📈📈📈] Autocomplete
- Old commands (also does not append to bash history)
- Directories
- Interactive
- [🥼 📈] Input. If a script requires further input midway of execution - it will simply hang
- [🥼 📈📈] Interrupt. Control-C will not halt the program. Impossible to intercept keystrokes, can bind a command
- [🥼🥼🥼🥼 📈] Any terminal control like coloring output, clearning lines
- This might be a deal breaker because buck for example seems to want to have an interactive terminal. Maybe if it cant detect a TTY to which its connected it just craps out 😢
- [🥼 📈📈] Visual appeal, usability
- Initialized with markdown and prefix command prompts with # for easier folding
- Other things like wrapping command that has ran in
backticks
- [🥼🥼 📈📈] Debugger integration
- Instead of creating my own shell, use an existing shell. Ideally it will provide filepath completions and things of that nature automatically. I found this cold one. brew install nushell
- To get it to work I need to create a tty, send input to the tell a typewriter, interpreting output will be more difficult, as I will need to process special sequences. For example clearing the line, clearing the entire screen etc.
- Before jumping into this new shell, let's try to get the old set up working
- Testing out the new shell should also be possible without changing how the extension works
- Now that I can ship binary with the extension, I can use node-tty. It uses node-gyp under the hood.
- [🥼🥼 📈📈] Refactor
- Flatten the god function
- Starting hard to read
- Need to add a separate command for lldb / bash
Log
November 23, blocking, nushell, colored
https://www.lihaoyi.com/post/BuildyourownCommandLinewithANSIescapecodes.html
https://stackoverflow.com/questions/36929209/read-ansi-escape-from-terminal
Instead of dealing with this escape sequence nonsense let's use xterm.js, headless version. Looks like it's used for almost this.
While reading the documentation I noticed UTF-16 for strings. Maybe I should use this to decode bite stream received from shell.
Clearly xterm.js has good potential to help with escape sequences https://github.com/xtermjs/xterm.js/blob/master/typings/xterm-headless.d.ts#L1072.
[Key] Documentation https://xtermjs.org/docs/guides/hooks/
Potential solution:
- Keep current one off command running
- Only use xterm.js headless as output. Create in new instance every time
- Use terminal right function to pipe text into terminal
- Get terminal buffer https://github.com/xtermjs/xterm.js/blob/master/typings/xterm-headless.d.ts#L1072
- Hypotheses: I can get un formated string (or formated structured string ranges) and simply print to vscode (optionally include annotations)
- From buffer get all lines, from each line get string, hopefully plain (or worst case scenario get cell. This actually has information about bold, color etcetera) https://github.com/xtermjs/xterm.js/blob/4.14.1/typings/xterm.d.ts#LL1338-L1338C26
- Once I have the current buffer, I can diff it with the previous buffer (should be synchronized with what vscode editor has starting from command output mark)
- Simply override range
- This should support fancy commandline applications that update in line, for example loading bars
Interactivity is still unsolved. Solving that by hand is definitely not great. Most likely I will still need to use xterm.js. Ideally I will subscribe to buffer change, or whatever event to be notified when application is asking for data (y/n), or input (for example arrow keys)
This will also take care of actually handling the command sequences. For example clearing a line.
Modifying algorithm outlined above
The question is is at worth my time right now?
Is tell a typewriter more important?)
- This will allow support for nushell, single line commands are not interactive
